Feature: don't reload prometheus if the scaled comp is itself

components/cluster/command/patch.go

@@ -46,7 +46,7 @@ func newPatchCmd() *cobra.Command {
 
 	cmd.Flags().BoolVar(&overwrite, "overwrite", false, "Use this package in the future scale-out operations")
 	cmd.Flags().StringSliceVarP(&gOpt.Nodes, "node", "N", nil, "Specify the nodes")
-	cmd.Flags().StringSliceVarP(&gOpt.Roles, "role", "R", nil, "Specify the role")
+	cmd.Flags().StringSliceVarP(&gOpt.Roles, "role", "R", nil, "Specify the roles")
 	cmd.Flags().Uint64Var(&gOpt.APITimeout, "transfer-timeout", 300, "Timeout in seconds when transferring PD and TiKV store leaders")
 	return cmd
 }

components/dm/command/patch.go

@@ -46,6 +46,6 @@ func newPatchCmd() *cobra.Command {
 
 	cmd.Flags().BoolVar(&overwrite, "overwrite", false, "Use this package in the future scale-out operations")
 	cmd.Flags().StringSliceVarP(&gOpt.Nodes, "node", "N", nil, "Specify the nodes")
-	cmd.Flags().StringSliceVarP(&gOpt.Roles, "role", "R", nil, "Specify the role")
+	cmd.Flags().StringSliceVarP(&gOpt.Roles, "role", "R", nil, "Specify the roles")
 	return cmd
 }

pkg/cluster/manager/builder.go

@@ -30,8 +30,8 @@ import (
 	"github.com/pingcap/tiup/pkg/set"
 )
 
-// Dynamic reload Prometheus configuration
-func buildDynReloadPromTasks(topo spec.Topology) []task.Task {
+// buildReloadPromTasks reloads Prometheus configuration
+func buildReloadPromTasks(topo spec.Topology, nodes ...string) []task.Task {
 	monitor := spec.FindComponent(topo, spec.ComponentPrometheus)
 	if monitor == nil {
 		return nil
@@ -40,11 +40,18 @@ func buildDynReloadPromTasks(topo spec.Topology) []task.Task {
 	if len(instances) == 0 {
 		return nil
 	}
-	var dynReloadTasks []task.Task
+	var tasks []task.Task
+	deletedNodes := set.NewStringSet(nodes...)
 	for _, inst := range monitor.Instances() {
-		dynReloadTasks = append(dynReloadTasks, task.NewBuilder().SystemCtl(inst.GetHost(), inst.ServiceName(), "reload", true).Build())
+		if deletedNodes.Exist(inst.ID()) {
+			continue
+		}
+		t := task.NewBuilder().
+			SystemCtl(inst.GetHost(), inst.ServiceName(), "reload", true).
+			Build()
+		tasks = append(tasks, t)
 	}
-	return dynReloadTasks
+	return tasks
 }
 
 func buildScaleOutTask(
@@ -298,7 +305,7 @@ func buildScaleOutTask(
 			return operator.Start(ctx, newPart, operator.Options{OptTimeout: gOpt.OptTimeout}, tlsCfg)
 		}).
 		Parallel(false, refreshConfigTasks...).
-		Parallel(false, buildDynReloadPromTasks(metadata.GetTopology())...)
+		Parallel(false, buildReloadPromTasks(metadata.GetTopology())...)
 
 	if final != nil {
 		final(builder, name, metadata)

pkg/cluster/manager/destroy.go

@@ -133,7 +133,10 @@ func (m *Manager) DestroyTombstone(
 		UpdateTopology(name, m.specManager.Path(name), clusterMeta, nodes)
 
 	regenConfigTasks, _ := buildRegenConfigTasks(m, name, topo, base, nodes)
-	t := b.ParallelStep("+ Refresh instance configs", true, regenConfigTasks...).Parallel(true, buildDynReloadPromTasks(metadata.GetTopology())...).Build()
+	t := b.
+		ParallelStep("+ Refresh instance configs", true, regenConfigTasks...).
+		Parallel(true, buildReloadPromTasks(metadata.GetTopology())...).
+		Build()
 	if err := t.Execute(task.NewContext()); err != nil {
 		if errorx.Cast(err) != nil {
 			// FIXME: Map possible task errors and give suggestions.

pkg/cluster/manager/scale_in.go

@@ -92,7 +92,7 @@ func (m *Manager) ScaleIn(
 
 	t := b.
 		ParallelStep("+ Refresh instance configs", force, regenConfigTasks...).
-		Parallel(force, buildDynReloadPromTasks(metadata.GetTopology())...).
+		Parallel(force, buildReloadPromTasks(metadata.GetTopology(), nodes...)...).
 		Build()
 
 	if err := t.Execute(task.NewContext()); err != nil {

pkg/cluster/task/builder.go

@@ -112,7 +112,7 @@ func (b *Builder) UpdateMeta(cluster string, metadata *spec.ClusterMeta, deleted
 	b.tasks = append(b.tasks, &UpdateMeta{
 		cluster:        cluster,
 		metadata:       metadata,
-		deletedNodesID: deletedNodeIds,
+		deletedNodeIDs: deletedNodeIds,
 	})
 	return b
 }
@@ -123,7 +123,7 @@ func (b *Builder) UpdateTopology(cluster, profile string, metadata *spec.Cluster
 		metadata:       metadata,
 		cluster:        cluster,
 		profileDir:     profile,
-		deletedNodesID: deletedNodeIds,
+		deletedNodeIDs: deletedNodeIds,
 	})
 	return b
 }

pkg/cluster/task/update_meta.go

@@ -25,15 +25,15 @@ import (
 type UpdateMeta struct {
 	cluster        string
 	metadata       *spec.ClusterMeta
-	deletedNodesID []string
+	deletedNodeIDs []string
 }
 
 // Execute implements the Task interface
 // the metadata especially the topology is in wide use,
 // the other callers point to this field by a pointer,
 // so we should update the original topology directly, and don't make a copy
 func (u *UpdateMeta) Execute(ctx *Context) error {
-	deleted := set.NewStringSet(u.deletedNodesID...)
+	deleted := set.NewStringSet(u.deletedNodeIDs...)
 	topo := u.metadata.Topology
 
 	tidbServers := make([]spec.TiDBSpec, 0)
@@ -154,5 +154,5 @@ func (u *UpdateMeta) Rollback(ctx *Context) error {
 
 // String implements the fmt.Stringer interface
 func (u *UpdateMeta) String() string {
-	return fmt.Sprintf("UpdateMeta: cluster=%s, deleted=`'%s'`", u.cluster, strings.Join(u.deletedNodesID, "','"))
+	return fmt.Sprintf("UpdateMeta: cluster=%s, deleted=`'%s'`", u.cluster, strings.Join(u.deletedNodeIDs, "','"))
 }

pkg/cluster/task/update_topology.go

@@ -17,7 +17,7 @@ type UpdateTopology struct {
 	cluster        string
 	profileDir     string
 	metadata       *spec.ClusterMeta
-	deletedNodesID []string
+	deletedNodeIDs []string
 }
 
 // String implements the fmt.Stringer interface
@@ -41,7 +41,7 @@ func (u *UpdateTopology) Execute(ctx *Context) error {
 
 	topo := u.metadata.Topology
 
-	deleted := set.NewStringSet(u.deletedNodesID...)
+	deleted := set.NewStringSet(u.deletedNodeIDs...)
 
 	var ops []clientv3.Op
 	var instances []spec.Instance

tests/tiup-cluster/script/scale_tools.sh

@@ -36,8 +36,12 @@ function scale_tools() {
 
     if [ $test_tls = true ]; then
         total_sub_one=18
+        total=19
+        total_add_one=20
     else
         total_sub_one=21
+        total=22
+        total_add_one=23
     fi
 
     echo "start scale in pump"
@@ -70,6 +74,14 @@ function scale_tools() {
     topo=./topo/full_scale_in_grafana.yaml
     tiup-cluster $client --yes scale-out $name $topo
 
+    echo "start scale out prometheus"
+    topo=./topo/full_scale_in_prometheus.yaml
+    wait_instance_num_reach $name $total_add_one $native_ssh
+    tiup-cluster $client --yes scale-out $name $topo
+    echo "start scale in prometheus"
+    tiup-cluster $client --yes scale-in $name -N n2:9090
+    wait_instance_num_reach $name $total $native_ssh
+
     # make sure grafana dashboards has been set to default (since the full_sale_in_grafana.yaml didn't provide a local dashboards dir)
     ! tiup-cluster $client exec $name -N n1 --command "grep magic-string-for-test /home/tidb/deploy/grafana-3000/dashboards/tidb.json"
 

tests/tiup-cluster/topo/full_scale_in_prometheus.yaml

@@ -0,0 +1,3 @@
+monitoring_servers:
+  - host: n2
+    rule_dir: /tmp/local/prometheus